gpu: nvgpu: add gpc_mask to gr/config unit

We get gpc_mask by calling GR HAL g->ops.gr.get_gpc_mask()

But gpc_mask should be logically owned by gr/config unit
Hence add new gpc_mask field to nvgpu_gr_config

Initialize it in nvgpu_gr_config_init() by calling a new HAL
g->ops.gr.config.get_gpc_mask() if available
If HAL is not defined we just initialize it based on gpc_count

Expose new API nvgpu_gr_config_get_gpc_mask() to get gpc_mask
and use this API now

Remove gr_gm20b_get_gpc_mask() and HAL g->ops.gr.get_gpc_mask()

Update GV100 and TU104 chip HALs to remove old and add new HAL

Add gpc_mask to struct tegra_vgpu_constants_params to support this
on vGPU. Also get gpc_mask from vGPU private data in
vgpu_gr_init_gr_config()

Jira NVGPU-1879

Change-Id: Ibdc89ea51df944dc7085920509e3536a5721efc0
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2016084
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/common/gr/config/gr_config.c

@@ -51,6 +51,12 @@ struct nvgpu_gr_config *nvgpu_gr_config_init(struct gk20a *g)
 		goto clean_up;
 	}
 
+	if (g->ops.gr.config.get_gpc_mask != NULL) {
+		config->gpc_mask = g->ops.gr.config.get_gpc_mask(g, config);
+	} else {
+		config->gpc_mask = BIT32(config->gpc_count) - 1;
+	}
+
 	config->pe_count_per_gpc = nvgpu_get_litter_value(g,
 		GPU_LIT_NUM_PES_PER_GPC);
 	if (config->pe_count_per_gpc > GK20A_GR_MAX_PES_PER_GPC) {
@@ -531,3 +537,8 @@ u32 nvgpu_gr_config_get_pes_tpc_mask(struct nvgpu_gr_config *config,
 {
 	return config->pes_tpc_mask[pes_index][gpc_index];
 }
+
+u32 nvgpu_gr_config_get_gpc_mask(struct nvgpu_gr_config *config)
+{
+	return config->gpc_mask;
+}

drivers/gpu/nvgpu/common/gr/config/gr_config_gm20b.c

@@ -77,3 +77,20 @@ u32 gm20b_gr_config_get_pd_dist_skip_table_size(void)
 {
 	return gr_pd_dist_skip_table__size_1_v();
 }
+
+u32 gm20b_gr_config_get_gpc_mask(struct gk20a *g,
+	struct nvgpu_gr_config *config)
+{
+	u32 val;
+
+	/*
+	 * For register NV_FUSE_STATUS_OPT_GPC a set bit with index i indicates
+	 * corresponding GPC is floorswept
+	 * But for s/w mask a set bit means GPC is enabled and it is disabled
+	 * otherwise
+	 * Hence toggle the bits of register value to get s/w mask
+	 */
+	val = g->ops.fuse.fuse_status_opt_gpc(g);
+
+	return (~val) & (BIT32(config->max_gpc_count) - 1U);
+}

drivers/gpu/nvgpu/common/gr/config/gr_config_gm20b.h

@@ -37,5 +37,7 @@ u32 gm20b_gr_config_get_zcull_count_in_gpc(struct gk20a *g,
 u32 gm20b_gr_config_get_pes_tpc_mask(struct gk20a *g,
 	struct nvgpu_gr_config *config, u32 gpc_index, u32 pes_index);
 u32 gm20b_gr_config_get_pd_dist_skip_table_size(void);
+u32 gm20b_gr_config_get_gpc_mask(struct gk20a *g,
+	struct nvgpu_gr_config *config);
 
 #endif /* NVGPU_GR_CONFIG_GM20B_H */

drivers/gpu/nvgpu/gm20b/gr_gm20b.c

@@ -558,23 +558,6 @@ void gr_gm20b_get_sm_dsm_perf_ctrl_regs(struct gk20a *g,
 	    g->ops.gr.ctxsw_prog.hw_get_perf_counter_control_register_stride();
 }
 
-u32 gr_gm20b_get_gpc_mask(struct gk20a *g)
-{
-	u32 val;
-	struct gr_gk20a *gr = &g->gr;
-
-	/*
-	 * For register NV_FUSE_STATUS_OPT_GPC a set bit with index i indicates
-	 * corresponding GPC is floorswept
-	 * But for s/w mask a set bit means GPC is enabled and it is disabled
-	 * otherwise
-	 * Hence toggle the bits of register value to get s/w mask
-	 */
-	val = g->ops.fuse.fuse_status_opt_gpc(g);
-
-	return (~val) & (BIT32(nvgpu_gr_config_get_max_gpc_count(gr->config)) - 1U);
-}
-
 void gr_gm20b_set_gpc_tpc_mask(struct gk20a *g, u32 gpc_index)
 {
 	nvgpu_tegra_fuse_write_bypass(g, 0x1);

drivers/gpu/nvgpu/gm20b/gr_gm20b.h

@@ -80,7 +80,6 @@ void gr_gm20b_get_sm_dsm_perf_ctrl_regs(struct gk20a *g,
 					       u32 **sm_dsm_perf_ctrl_regs,
 					       u32 *ctrl_register_stride);
 void gr_gm20b_set_gpc_tpc_mask(struct gk20a *g, u32 gpc_index);
-u32 gr_gm20b_get_gpc_mask(struct gk20a *g);
 void gr_gm20b_load_tpc_mask(struct gk20a *g);
 void gr_gm20b_program_sm_id_numbering(struct gk20a *g,
 					     u32 gpc, u32 tpc, u32 smid);

drivers/gpu/nvgpu/gv100/hal_gv100.c

@@ -367,7 +367,6 @@ static const struct gpu_ops gv100_ops = {
 		.falcon_load_ucode = gr_gm20b_load_ctxsw_ucode_segments,
 		.load_ctxsw_ucode = gr_gm20b_load_ctxsw_ucode,
 		.set_gpc_tpc_mask = gr_gv100_set_gpc_tpc_mask,
-		.get_gpc_mask = gr_gm20b_get_gpc_mask,
 		.alloc_obj_ctx = gk20a_alloc_obj_ctx,
 		.bind_ctxsw_zcull = gr_gk20a_bind_ctxsw_zcull,
 		.get_zcull_info = gr_gk20a_get_zcull_info,
@@ -585,6 +584,7 @@ static const struct gpu_ops gv100_ops = {
 			.dump_ctxsw_stats = gp10b_ctxsw_prog_dump_ctxsw_stats,
 		},
 		.config = {
+			.get_gpc_mask = gm20b_gr_config_get_gpc_mask,
 			.get_gpc_tpc_mask = gm20b_gr_config_get_gpc_tpc_mask,
 			.get_tpc_count_in_gpc =
 				gm20b_gr_config_get_tpc_count_in_gpc,

drivers/gpu/nvgpu/include/nvgpu/gk20a.h

@@ -292,7 +292,6 @@ struct gpu_ops {
 				struct gk20a_ctxsw_ucode_segments *segments,
 				u32 reg_offset);
 		int (*load_ctxsw_ucode)(struct gk20a *g);
-		u32 (*get_gpc_mask)(struct gk20a *g);
 		void (*set_gpc_tpc_mask)(struct gk20a *g, u32 gpc_index);
 		int (*alloc_obj_ctx)(struct channel_gk20a  *c,
 				     u32 class_num, u32 flags);
@@ -619,6 +618,8 @@ struct gpu_ops {
 		} ctxsw_prog;
 
 		struct {
+			u32 (*get_gpc_mask)(struct gk20a *g,
+				struct nvgpu_gr_config *config);
 			u32 (*get_gpc_tpc_mask)(struct gk20a *g,
 				struct nvgpu_gr_config *config, u32 gpc_index);
 			u32 (*get_tpc_count_in_gpc)(struct gk20a *g,

drivers/gpu/nvgpu/include/nvgpu/gr/config.h

@@ -47,6 +47,7 @@ struct nvgpu_gr_config {
 	u32 *gpc_zcb_count;
 	u32 *pes_tpc_count[GK20A_GR_MAX_PES_PER_GPC];
 
+	u32 gpc_mask;
 	u32 *gpc_tpc_mask;
 	u32 *pes_tpc_mask[GK20A_GR_MAX_PES_PER_GPC];
 	u32 *gpc_skip_mask;
@@ -92,5 +93,6 @@ u32 nvgpu_gr_config_get_gpc_skip_mask(struct nvgpu_gr_config *config,
 	u32 gpc_index);
 u32 nvgpu_gr_config_get_pes_tpc_mask(struct nvgpu_gr_config *config,
 	u32 gpc_index, u32 pes_index);
+u32 nvgpu_gr_config_get_gpc_mask(struct nvgpu_gr_config *config);
 
 #endif /* NVGPU_GR_CONFIG_H */

drivers/gpu/nvgpu/include/nvgpu/vgpu/tegra_vgpu.h

@@ -501,6 +501,7 @@ struct tegra_vgpu_constants_params {
 	u32 sm_arch_warp_count;
 	u32 max_gpc_count;
 	u32 gpc_count;
+	u32 gpc_mask;
 	u32 max_tpc_per_gpc_count;
 	u32 num_fbps;
 	u32 fbp_en_mask;

drivers/gpu/nvgpu/os/linux/ioctl_ctrl.c

@@ -284,6 +284,7 @@ gk20a_ctrl_ioctl_gpu_characteristics(
 
 	gpu.num_gpc = nvgpu_gr_config_get_gpc_count(g->gr.config);
 	gpu.max_gpc_count = nvgpu_gr_config_get_max_gpc_count(g->gr.config);
+	gpu.gpc_mask = nvgpu_gr_config_get_gpc_mask(g->gr.config);
 
 	gpu.num_tpc_per_gpc = nvgpu_gr_config_get_max_tpc_per_gpc_count(g->gr.config);
 
@@ -291,12 +292,6 @@ gk20a_ctrl_ioctl_gpu_characteristics(
 
 	gpu.compression_page_size = g->ops.fb.compression_page_size(g);
 
-	if (g->ops.gr.get_gpc_mask) {
-		gpu.gpc_mask = g->ops.gr.get_gpc_mask(g);
-	} else {
-		gpu.gpc_mask = BIT32(gpu.num_gpc) - 1;
-	}
-
 	gpu.flags = nvgpu_ctrl_ioctl_gpu_characteristics_flags(g);
 
 	gpu.arch = g->params.gpu_arch;

drivers/gpu/nvgpu/tu104/hal_tu104.c

@@ -384,7 +384,6 @@ static const struct gpu_ops tu104_ops = {
 		.falcon_load_ucode = gr_gm20b_load_ctxsw_ucode_segments,
 		.load_ctxsw_ucode = gr_gm20b_load_ctxsw_ucode,
 		.set_gpc_tpc_mask = gr_gv100_set_gpc_tpc_mask,
-		.get_gpc_mask = gr_gm20b_get_gpc_mask,
 		.alloc_obj_ctx = gk20a_alloc_obj_ctx,
 		.bind_ctxsw_zcull = gr_gk20a_bind_ctxsw_zcull,
 		.get_zcull_info = gr_gk20a_get_zcull_info,
@@ -610,6 +609,7 @@ static const struct gpu_ops tu104_ops = {
 			.dump_ctxsw_stats = gp10b_ctxsw_prog_dump_ctxsw_stats,
 		},
 		.config = {
+			.get_gpc_mask = gm20b_gr_config_get_gpc_mask,
 			.get_gpc_tpc_mask = gm20b_gr_config_get_gpc_tpc_mask,
 			.get_tpc_count_in_gpc =
 				gm20b_gr_config_get_tpc_count_in_gpc,

drivers/gpu/nvgpu/vgpu/gr_vgpu.c

@@ -600,6 +600,7 @@ static int vgpu_gr_init_gr_config(struct gk20a *g, struct gr_gk20a *gr)
 
 	config->max_gpc_count = priv->constants.max_gpc_count;
 	config->gpc_count = priv->constants.gpc_count;
+	config->gpc_mask = priv->constants.gpc_mask;
 	config->max_tpc_per_gpc_count = priv->constants.max_tpc_per_gpc_count;
 
 	config->max_tpc_count = config->max_gpc_count * config->max_tpc_per_gpc_count;